When Is Funding Charged on a Bitcoin Perp? At the Stamp, and Dodging One Costs 8 Times More

TLDR. Perpetual funding is not rent that accrues while you hold. It is a transfer taken from whoever is open at the settlement stamp, so a position opened and closed between two stamps pays nothing at all, and a position held for ten minutes across one pays the full interval rate. That makes the interval the thing to read, and we measured it first hand on 2026-09-17 rather than quoting a help page: over the last 90 days Binance, Bybit, OKX and Bitget stamped the Bitcoin perpetual every 8.0 hours, 270 times each, with no gap off that clock by as much as a tenth of a second, while Hyperliquid stamped it every 1.0 hours, 2,160 times. Holding through all of them cost a $10,000 long $157.64 on Binance and $197.38 on Hyperliquid. The obvious trade is to close before each stamp, and it does not survive contact with the fee schedule: the biggest single settlement in the whole set was 0.0127 per cent of notional, a round trip at a 0.05 per cent taker fee costs 0.10 per cent, and dodging all 270 Binance stamps would have spent 27 per cent of notional in fees to avoid 1.58 per cent of funding. The stamp is worth timing only if you were closing anyway.

When is funding actually charged on a Bitcoin perpetual?

At the stamp, and only at the stamp. Binance, Bybit, OKX and Bitget each publish the rule in their own help pages, in almost the same words: you pay or receive funding only if you hold a position at the funding time, and a position closed before it is charged nothing. Gate.io is the one that never gives the unconditional version, not because it works differently but because its help page warns instead that a position opened at the settlement second "may still be charged or credited", which is not the same promise. You can see the clock itself in the data without reading a single help page. We pulled each venue's own public funding history for the Bitcoin perpetual and differenced consecutive settlement timestamps. On Bybit, OKX and Bitget all 269 gaps in 270 settlements measured 8.0 hours to the second, and on Binance every gap sat within 26 milliseconds of it, so the three daily stamps at 00:00, 08:00 and 16:00 UTC held without a single exception across 90 days. Hyperliquid runs a different clock entirely, 2,159 consecutive gaps of 1.0 hours, none of them more than 130 milliseconds off. Deribit does not use a stamp at all: it quotes an 8 hour rate but accrues it continuously against the position, so by its published time fraction formula a position held for one minute accrues exactly one four hundred and eightieth of the rate. The rule itself is not new here, and the Athenum explainer on perpetual futures states it plainly; what follows is the part nobody measures, which is what the rule is worth in money.

That difference is not a footnote for a minority of the market. It decides how much of the open interest can ever be timed.

Athenum's cross-exchange view at 2026-09-17 08:00 UTC: of $21.16B in Bitcoin perpetual open interest, $17.32B sits on an 8 hour stamp (Binance $8.277B, Bybit $4.272B, OKX $2.203B, Bitget $2.563B), $2.89B on an hourly stamp at Hyperliquid, and $0.80B at Deribit, which accrues funding continuously.

Read from Athenum's normalized cross-exchange feed for the hour stamped 2026-09-17 08:00 UTC, aggregate Bitcoin perpetual open interest was $21.16B. Of that, $17.32B sat on venues that settle every eight hours, made up of $8.277B on Binance, $4.272B on Bybit, $2.203B on OKX and $2.563B on Bitget. Hyperliquid carried $2.89B on the hourly clock and Deribit $0.80B with no stamp to time at all. Binance, Bybit, OKX, Bitget, Hyperliquid and Deribit together sum to $21.01B, which is 99.3 per cent of the aggregate figure, and the remainder sits on boards the feed tracks without breaking them out in this view. So over four fifths of that open interest sits behind a window of nearly eight hours in which a position is genuinely free of funding, Hyperliquid's window is under an hour, and on Deribit there is no window at all.

What did a Bitcoin long actually pay by holding through every stamp?

Between 1.06 and 1.97 per cent of notional over 90 days, depending on the venue, and the spread between boards is wider than most traders assume. We took the same 90 days of funding accrual on every board, ending at the settlement stamped 2026-09-17 08:00 UTC, which is 270 eight hour stamps or 2,160 hourly ones, and summed every settlement a long would have paid.

Cumulative funding on a $10,000 Bitcoin long held through 90 days of accrual to 2026-09-17 08:00 UTC: $197.38 on Hyperliquid, $157.64 on Binance, $132.22 on OKX, $122.47 on Bitget and $105.65 on Bybit.

On a constant $10,000 notional the bill was $197.38 on Hyperliquid, $157.64 on Binance, $132.22 on OKX, $122.47 on Bitget and $105.65 on Bybit. As annualized rates those are 8.00, 6.39, 5.36, 4.97 and 4.28 per cent. Two things that number is not. It is not a venue ranking: it is one window, and the Athenum post on the funding rate cap publishes a window where the order is different, with interquartile bands that overlap each other heavily. And it is not a bill anyone was actually sent, because each venue multiplies the rate by a notional it computes its own way, Binance and Bybit against the mark price and Hyperliquid against its spot oracle, so a real position that moved with the market pays a marked figure rather than a constant one. Gate.io is missing from this comparison on purpose: its public endpoint returned only the last 90 settlements, a 30 day window rather than 90 days, and a sum over a shorter window is not comparable. It is in the box plot, where the unit is a single settlement, and in the payments chart below on its own 30 day window, labelled as such. It is out of the cumulative chart and out of every 90 day sum quoted here.

The single settlement behind those totals is small and unusually repetitive. On Binance the median stamp was 0.00591 per cent of notional with a middle half from 0.00382 to 0.00820, and 38 of the 270 settlements printed exactly 0.01000 per cent: that is the interest rate component showing through whenever the premium is close to it, not a ceiling the market ran into. Binance publishes a funding cap of 0.3 per cent for this contract, thirty times the largest settlement it printed in the window. Hyperliquid shows the same fingerprint far more strongly, with 1,287 of its 2,160 hourly settlements printing exactly 0.00125 per cent, which is the same 0.01 per cent anchor divided across eight hours. A distribution with a point mass that large is not well described by its average, which is why the box plot published with this post reports a median and a middle half rather than a mean. The free Athenum funding rate calculator turns any of these per interval rates into an annual cost, and why funding intervals are set per symbol covers the boards where the interval is not eight hours at all.

Is it worth closing before the stamp to avoid funding?

Almost never, and the reason is arithmetic rather than opinion. Across all 3,330 settlements in this study, 90 days each on Binance, Bybit, OKX, Bitget and Hyperliquid and 30 on Gate.io, the largest single settlement in absolute terms was 0.0127 per cent of notional, and it was a negative print on Bitget, which is money paid to longs rather than taken from them. The most a long actually paid at any one stamp was 0.01000 per cent, the interest rate anchor, printed on Binance, Bybit, OKX, Bitget and Gate.io alike. The comparisons below still use the 0.0127, because testing the fee against the biggest settlement of either sign is the version of this test that is hardest on our own conclusion. The set also mixes single stamps from an hourly board with single stamps from eight hourly boards, so it is worth stating the other unit too: Hyperliquid's worst eight consecutive hourly stamps came to 0.0222 per cent, larger than any single eight hour print anywhere in the set and still four and a half times below the round trip. A round trip at the 0.05 per cent taker fee Binance and OKX publish for a regular user costs 0.10 per cent of notional, which is 7.9 times that largest settlement and 16.9 times the median Binance stamp. Zero of the 3,330 settlements reached the cost of the trade that would have avoided them. Bybit charges a regular user 0.055 per cent a side and Hyperliquid 0.045 per cent at its base tier, so the gap is wider on one board and narrower on another, and it never closes. Turn it around and the break even is the clearest way to see it: to make dodging worthwhile you would need to trade both legs for under 0.00635 per cent a side against the largest settlement in the set, and under 0.00295 per cent against the median Binance stamp. No published fee on any of these boards is that low. Even two maker fills at Binance's 0.02 per cent come to 0.04 per cent a round trip, which is 3.1 times the largest settlement and 6.8 times the median, and not one of the 3,330 reached even that.

Run it over the full window and the gap gets wider still. Closing before all 270 Binance stamps and re-entering after each one means 270 round trips, which at 0.05 per cent a side costs 27 per cent of notional, to avoid the 1.58 per cent of funding a long actually paid. That is seventeen times the cost of simply holding. The same trade on Hyperliquid would mean 2,160 round trips. The fee is only the part you can price in advance: being flat across a settlement instant also means 270 exits and re-entries at whatever the book offers, and the spread and the gap risk of those round trips are real costs the arithmetic above never charges the dodger.

Widen the window and the conclusion survives while the mood of the market does not. Binance's endpoint returns 1,000 settlements in one call, covering 2025-09-17 16:00 UTC to 2026-08-16 16:00 UTC, and 259 of those 1,000 were negative, 25.9 per cent, against 5 of 270 in the 90 days measured above. That call stops a month before this study was captured, so it overlaps the 90 days rather than standing apart from them: read it as a longer sample, not an independent one. The recent window is still unusually one-sided and should not be read as normal. The largest single settlement in all 1,000 was 0.01518 per cent of notional in absolute terms, again a negative print, still 6.6 times smaller than the round trip that would have avoided it, and not one of the 1,000 reached it either.

Three honest qualifications. First, even that longer window is calm by the standards of a squeeze, when a single settlement can be far larger, so check the live rate before you generalise. Second, the stamp is a window rather than an instant: Binance documents a 15 second deviation in when the fee actually lands, OKX up to a minute, and Bybit says outright that opening or closing within 5 seconds of the funding time does not guarantee which side of it you land on, with no reimbursement if you guess wrong. Cutting it fine is a coin flip against a counterparty that has told you it will not refund. Third, the arithmetic changes if you are closing anyway: if a trade is ending within the hour before a stamp, moving the exit ten minutes earlier costs nothing and saves a full interval. That is the only version of this idea that pays, and it is a scheduling decision rather than a strategy.

What does a stamp dodger give up when funding turns negative?

The receipts, and how much that costs is a venue property rather than a market one. Funding is symmetric in direction: when the rate is negative, shorts pay longs, and a long who is flat at the stamp collects nothing. It is not symmetric in size, because the interest rate component gives the series a positive drift, so a dodger really does skip more than they forgo.

Skipping every stamp skips both sides: over the window to 2026-09-17 Binance longs paid 1.582 per cent of notional and were paid 0.006 per cent back across 5 negative stamps of 270, while Bitget longs paid 1.375 per cent and collected 0.150 per cent across 40 negative stamps of 270.

How much a dodger forgoes depends entirely on the board. On Binance only 5 of 270 settlements were negative and the receipts came to 0.006 per cent of notional, so skipping them costs a rounding error. On Bitget 40 of 270 were negative and the receipts came to 0.150 per cent, about a tenth of what the same position paid out. Bybit sat between them at 36 negative stamps and 0.085 per cent, OKX at 24 and 0.047 per cent, and Hyperliquid at 219 of 2,160 hourly stamps and 0.108 per cent. The venue with the cheapest net funding in this window, Bybit, got there partly by paying its longs more often than Binance did, not only by charging less. A negative funding rate is rare on Bitcoin but not evenly rare across venues, and the PnL calculator will carry the funding leg in both directions if you want to price a specific hold. What the rate does in the hours leading up to the stamp is a separate question with its own answer in funding before settlement.

How would you reproduce this?

Public endpoints, no key, and one subtraction. Everything above came from each venue's own funding history plus Athenum's cross-exchange feed, captured on 2026-09-17:

1. Pull the funding history for the Bitcoin perpetual from each board: Binance fapi/v1/fundingRate for BTCUSDT, Bybit v5/market/funding/history for the linear BTCUSDT, OKX public/funding-rate-history for BTC-USDT-SWAP, Hyperliquid's info endpoint with type: fundingHistory, Bitget's history-fund-rate and Gate.io's futures/usdt/funding_rate. Each is paginated differently and only Binance returns a full 90 days in one call.

2. Window every series to the same start and end instant before you compare anything. Gate.io will not reach back 90 days on this route, which is why its window sums are kept out of every 90 day comparison here and carry a 30 day label wherever they do appear.

3. Measure the interval rather than assuming it: difference consecutive settlement timestamps and take the median, then check the minimum and maximum too. If those three numbers are not identical, the board changed its clock inside your window and your per day figures are wrong. This is not hypothetical. Binance has switched a contract from eight hourly to hourly settlement when the previous rate hit its cap since May 2025, OKX added a stepped version of the same escalation in April 2026, and Bybit reserves the right to change a settlement frequency without a separate announcement on the contracts that scheme covers.

4. Sum the rates for the cost of holding, and sum the positive and negative settlements separately for what a dodger would skip on each side. Report the count of negative stamps next to the sum, because 0.006 per cent spread over 5 settlements and 0.150 per cent spread over 40 are different facts about the same market.

5. Compare a single settlement against your own fee schedule, not against the standard rate. The 0.10 per cent round trip used here is two taker fills at the published 0.05 per cent; a maker fill, a fee tier or a rebate moves the break even and is the one input a reader should replace with their own.

One caution about quoting any of these figures back later. The interval is a structural fact that moves rarely, but the rates are live and the window slides: every number in this post describes the 90 days ending 2026-09-17 08:00 UTC and nothing else. Re-run the pull rather than citing the table.

Every figure here came off public endpoints that need no key, plus Athenum's own normalized feed, which reads Binance, Bybit, OKX, Bitget, Deribit and Hyperliquid side by side for exactly this kind of comparison. Nothing about the 34 calculators is gated: every one of them is free, none asks for an account or an email address, and none sets a limit on how many times you run it. If you would rather watch funding across those boards than paginate each venue's endpoint by hand, a free 7 day Pro+ trial opens the terminal and asks for no card.